Big Data Analytics on Object Storage -- Hadoop over Ceph* Object Storage with SSD Cache

Transcription

1 Big Data Analytics on Object Storage -- Hadoop over Ceph* Object Storage with SSD Cache David Cohen ) Yuan Zhou Jun Sun Weiting Chen Sep 2015

2 Agenda Big Data Analytics over Cloud Deployment considerations Design details of BDA over Ceph* Object Storage Sample Performance testing and results 2

3 Introduction Intel Cloud computing and Big Data Engineering Team Global team, local focus Open Spark, Hadoop, OpenStack, Ceph, NoSQL etc. Working with community and end customers closely Technology and Innovation oriented Real-time, in-memory, complex analytics Structure and unstructured data Agility, Multitenancy, Scalability and elasticity Bridging advanced research and real-world applications 3

4 Agenda Big Data Analytics over Cloud Deployment considerations Design details of BDA over Ceph* Object Storage Sample Performance testing and results 4

5 Big Data Analytics over Cloud You or someone at your company is using AWS, Azure, or Google cloud platform You re probably doing it for easy access to OS instances, but also the modern application features, e.g. AWS EMR or RDS or Storage Migrating to, or even using, OpenStack infrastructure for workloads means having application features, e.g. Sahara & Trove Writing applications is complex enough without having to manage supporting (non-value-add) infrastructure Things are also happening on the cloud computing side 5

6 Big Data Analytics over Cloud: OpenStack Sahara Repeatable cluster provisioning and management operations Data processing workflows (EDP) Cluster scaling (elasticity), Storage integration (Swift, Cinder, HCFS) Network and security group (firewall) integration Service anti-affinity (fault domains & efficiency) 6

7 Big data analytics on Object Storage Object storage provides restful/http access which is a good choice for archival storage It s just like a K-V storage and be able to build on commodity hardware Currently Hadoop solution usually relies on HDFS Hadoop over Swift (SwiftFS) provides an much easier way Which saves much efforts on the copying between HDFS and local storage 7

8 Agenda Big Data Analytics over Cloud Deployment considerations Design details of BDA over Ceph* Object Storage Sample Performance testing and results 8

9 Deployment Consideration Matrix Data Processing API Traditional EDP (Sahara native) 3rd party APIs Distro/Plugin Vanilla Spark Storm CDH HDP MapR Compute VM Container Bare-metal Storage Tenant vs. Admin provisioned Disaggregated vs. Collocated HDFS vs. other options 9

10 Storage Architecture Tenant provisioned (in VM) HDFS in the same VMs of computing tasks vs. in the different VMs Ephemeral disk vs. Cinder volume #1 Computing Task HDF S VM #2 #3 #4 Computing Task VM Computing Task VM HDF S VM Computing Task VM Computing Task VM Admin provided Logically disaggregated from computing tasks Physical collocation is a matter of deployment For network remote storage, Neutron DVR is very useful feature A disaggregated (and centralized) storage system has significant values No data silos, more business opportunities Could leverage Manila service Allow to create advanced solutions (.e.g. inmemory overlayer) More vendor specific optimization opportunities Legacy NFS Host HDFS GlusterFS Host HDFS Ceph* External HDFS Host HDFS Scenario #1: computing and data service collocate in the VMs Scenario #2: data service locates in the host world Scenario #3: data service locates in a separate VM world Scenario #4: data service locates in the remote network Swift 10

11 Agenda Big Data Analytics over Cloud Deployment considerations Design details of BDA over Ceph* Object Storage Sample Performance testing and results 11

12 Big data analytics on Ceph Object Storage Ceph provides a unified storage solution, which saves more man power to maintain another different setup for storage. We plan to build a reference solution on Hadoop over multiple Ceph* with SSD cache, similar with Hadoop over Swift. In this solution there s a requirement that all the storage servers are in a isolated network with the Hadoop cluster. The instances will play as the connectors of these two networks. We'll leverage Ceph* Cache Tier technology to cache the data in each servers. The reason for not using CephFS: CephFS is not so mature comparing with Existing deployment in DC may require a isolated network between storage and compute nodes 11

13 Servic e Isolated network Ceph* RADO S Scheduler FS 1 - Proxy (NEW!) MON Gateway 2 Server 1 M/R FS 3 (modified) OSD(SSD ) Rack 1 4 OSD(SSD ) Rack 2 Server 2 M/R FS (modified) OSD OSD OSD OSD OSD OSD Gateway OSD(SSD ) OSD(SSD ) Cache Tier Base Tier Ceph* with SSD cache 1. Scheduler ask service where a particular block locates (control path) -Proxy returns the closest active instance(s) 2. Scheduler allocates a task on the server that is near to the data 3. Task access data from nearby (data path) 4. get/put data from the CT, and CT would get/put data from BT if necessary (data path) 12

14 FS a new adaptor for HCFS Scheduler :// FS RADOS New filesystem URL rgw:// 1. Forked from Hadoop-8545(SwiftFS) 2. Hadoop is able to talk to a cluster with this plugin 3. A new block concept was added since Swift doesn t support blocks file1 Thus scheduler could use multiple tasks to access the same file Based on the location, FS is able to read from the closest through range GET API

15 -Proxy Give out the closest instance -Proxy (NEW!) Scheduler :// FS RADOS file1 1. Before get/put, FS would try to get the location of each block from -Proxy 1. One topology file of the cluster is generated 2. -Proxy would get the manifest from the head object first(librados + getxattr) 3. Then based on the crushmap -proxy can get the location of each object block(ceph osd map) 4. -proxy could get the closest the data osd info and the topology file(simple lookup in the topology file)

16 Serve the data requests 1. Setting up on a Cache Tier thus we could use SSD as the cache. Scheduler FS 1. With some dedicated chunk size: e.g., 64MB considering the data are quite big usually 2. rgw_max_chunk_size, rgw_obj_strip_size 2. Based on the account/container/object name, could get/put the content. 1. Using Range Read to get each chunk 3. We ll use write-through mode here as a start point to bypass the data consistency issue. Service Ceph* RADOS - Proxy (NEW!) MON Gateway (modified) OSD OSD (modified) OSD OSD OSD OSD OSD OSD Gateway OSD OSD Cache Tier Base Tier 15

17 Rack 1 Server 1 Rack 2 Server 2 Sample Get(mapper) Servic e Ceph* RADO S Scheduler FS 1 - Proxy (NEW!) MON Gateway 2 M/R FS 3 (modified) OSD(SSD ) 4 OSD(SSD ) M/R FS (modified) OSD OSD OSD OSD OSD OSD Gateway OSD(SSD ) OSD(SSD ) Cach e Tier Base Tier Scheduler asks -Proxy for the location of a big data file -Proxy looks for this file in a specified container (configured by operator) with a HEAD request to then returns the RADOS objects info in the manifest -Proxy looks for these objects in RADOS using Crush Get the target OSD in CT for each object. And with the OSD info, we know which instance is the closest. -Proxy would also monitors these instances and do the failover by returning the active instances Scheduler allocates a task on the server that is near to the data( instance) FS would issue a range read to get the block Ceph* CT would automatically handle the consistency here. 16

18 Service Ceph* RADO S Scheduler FS 1 - Proxy (NEW!) MON Gateway 2 Server 1 M/R FS 3 (modified) OSD(SSD ) Rack 1 4 OSD(SSD ) Rack 2 Server 2 M/R FS (modified) OSD OSD OSD OSD OSD OSD Gateway OSD(SSD ) OSD(SSD ) Cache Tier Base Tier Sample Put(reducer) Scheduler asks -Proxy for the location of a big data file -Proxy monitors the instances and returns an active instances list, with the closest instance at the first place and several other instances e.g., [ip1:/a/c/o(same rack), ip2:/a/c/o, ip3:/a/c/o] Reducer writes the output to would split the object into multiple RADOS objects Ceph* CT would automatically handle the consistency here. 17

19 Agenda Big Data Analytics over Cloud Deployment considerations Design details of BDA over Ceph* Sample Performance testing and results 19

20 Performance testing results VM HDFS.. VM HDFS VM VM Host Nova Inst. Store.. JBOD vs... Host Ceph/Swift JBOD o FS is Better than SwiftFS but worse than HDFS o Big gap with HDFS o Profiling data shows the biggest overhead comes from rename o Object stores use copy + delete, which is quite heavy 19

21 Rename in Reduce Task The output of the reduce function is written to a temporary location in HDFS. After completing, the output will automatically renamed from its temporary location to its final location. Object storage cannot support rename, copy and delete are used for rename function. HDFS Rename -> Change METADATA in Name Node Swift Rename -> Copy new object and Delete the older one Rename -> Copy new header file and Delete the old header file

22 Summary Big data analytics running on cloud will ease your work However each deployment has Pros and Cons Performance penalty is big, optimization needed Comparing with local HDFS setup, performance is 50% worse. The biggest gap is on the rename part. Intel is optimizing CEPH for Big data world 21

23 Next Step Finish the development(80% done) and complete the performance testing work Open source code repo(tbd) Optimize the rename part 23

24 Q&A 23

25 Legal Notices and Disclaimers Intel technologies features and benefits depend on system configuration and may require enabled hardware, software or service activation. Learn more at intel.com, or from the OEM or retailer. No computer system can be absolutely secure. Tests document performance of components on a particular test, in specific systems. Differences in hardware, software, or configuration will affect actual performance. Consult other sources of information to evaluate performance as you consider your purchase. For more complete information about performance and benchmark results, visit Cost reduction scenarios described are intended as examples of how a given Intel-based product, in the specified circumstances and configurations, may affect future costs and provide cost savings. Circumstances will vary. Intel does not guarantee any costs or cost reduction. This document contains information on products, services and/or processes in development. All information provided here is subject to change without notice. Contact your Intel representative to obtain the latest forecast, schedule, specifications and roadmaps. Statements in this document that refer to Intel s plans and expectations for the quarter, the year, and the future, are forward-looking statements that involve a number of risks and uncertainties. A detailed discussion of the factors that could affect Intel s results and plans is included in Intel s SEC filings, including the annual report on Form 10-K. The products described may contain design defects or errors known as errata which may cause the product to deviate from published specifications. Current characterized errata are available on request. No license (express or implied, by estoppel or otherwise) to any intellectual property rights is granted by this document. Intel does not control or audit third-party benchmark data or the web sites referenced in this document. You should visit the referenced web site and confirm whether referenced data are accurate. Intel, Xeon and the Intel logo are trademarks of Intel Corporation in the United States and other countries. *Other names and brands may be claimed as the property of others Intel Corporation. 24

26 Legal Information: Benchmark and Performance Claims Disclaimers Software and workloads used in performance tests may have been optimized for performance only on Intel microprocessors. Performance tests, such as SYSmark* and MobileMark*, are measured using specific computer systems, components, software, operations and functions. Any change to any of those factors may cause the results to vary. You should consult other information and performance tests to assist you in fully evaluating your contemplated purchases, including the performance of that product when combined with other products. Tests document performance of components on a particular test, in specific systems. Differences in hardware, software, or configuration will affect actual performance. Consult other sources of information to evaluate performance as you consider your purchase. Test and System Configurations: See Back up for details. For more complete information about performance and benchmark results, visit 25

27 Risk Factors The above statements and any others in this document that refer to plans and expectations for the first quarter, the year and the future are forwardlooking statements that involve a number of risks and uncertainties. Words such as "anticipates," "expects," "intends," "plans," "believes," "seeks," "estimates," "may," "will," "should" and their variations identify forward-looking statements. Statements that refer to or are based on projections, uncertain events or assumptions also identify forward-looking statements. Many factors could affect Intel's actual results, and variances from Intel's current expectations regarding such factors could cause actual results to differ materially from those expressed in these forward-looking statements. Intel presently considers the following to be important factors that could cause actual results to differ materially from the company's expectations. Demand for Intel s products is highly variable and could differ from expectations due to factors including changes in the business and economic conditions; consumer confidence or income levels; customer acceptance of Intel s and competitors products; competitive and pricing pressures, including actions taken by competitors; supply constraints and other disruptions affecting customers; changes in customer order patterns including order cancellations; and changes in the level of inventory at customers. Intel s gross margin percentage could vary significantly from expectations based on capacity utilization; variations in inventory valuation, including variations related to the timing of qualifying products for sale; changes in revenue levels; segment product mix; the timing and execution of the manufacturing ramp and associated costs; excess or obsolete inventory; changes in unit costs; defects or disruptions in the supply of materials or resources; and product manufacturing quality/yields. Variations in gross margin may also be caused by the timing of Intel product introductions and related expenses, including marketing expenses, and Intel s ability to respond quickly to technological developments and to introduce new features into existing products, which may result in restructuring and asset impairment charges. Intel's results could be affected by adverse economic, social, political and physical/infrastructure conditions in countries where Intel, its customers or its suppliers operate, including military conflict and other security risks, natural disasters, infrastructure disruptions, health concerns and fluctuations in currency exchange rates. Results may also be affected by the formal or informal imposition by countries of new or revised export and/or import and doing-business regulations, which could be changed without prior notice. Intel operates in highly competitive industries and its operations have high costs that are either fixed or difficult to reduce in the short term. The amount, timing and execution of Intel s stock repurchase program and dividend program could be affected by changes in Intel s priorities for the use of cash, such as operational spending, capital spending, acquisitions, and as a result of changes to Intel s cash flows and changes in tax laws. Product defects or errata (deviations from published specifications) may adversely impact our expenses, revenues and reputation. Intel s results could be affected by litigation or regulatory matters involving intellectual property, stockholder, consumer, antitrust, disclosure and other issues. An unfavorable ruling could include monetary damages or an injunction prohibiting Intel from manufacturing or selling one or more products, precluding particular business practices, impacting Intel s ability to design its products, or requiring other remedies such as compulsory licensing of intellectual property. Intel s results may be affected by the timing of closing of acquisitions, divestitures and other significant transactions. A detailed discussion of these and other factors that could affect Intel s results is included in Intel s SEC filings, including the company s most recent reports on Form 10-Q, Form 10-K and earnings release. 26 Rev. 1/15/15

28 Backup 28

29 MOC Project OpenStack.pdf

30 HDFS Server 1 Server 2 Scheduler 2 M/R M/R Scheduler ask HDFS service where a particular block locates (control path) HDFS Adapter 1 HDFS Adapter 3 HDFS Adapter Scheduler allocates a task on the server that is near to the data Task access data from local (data path) HDFS Service Name Node Data Node 4 Data Node In case of write, HDFS replicates the data autonomously (data path)

31 Rack 1 Server 1 Rack 2 Server 2 Swift Swift Servic e Scheduler Swift Adapter 1 Proxy Node M/R Swift Adapter Storage Node M/R Swift Adapter Storage Node 1. Scheduler ask Swift service where a particular block locates (control path) Thru regular read API The Swift Proxy service needs to be specially configured so that it will handle the read API in an unusual way return a list of locations instead of data. 2. Scheduler allocates a task on the server that is near to the data 3. Task access data from nearby (data path) 4. In case of write, Proxy will take the responsibility to write all the replicas (data path)

32 Servic e Ceph* RADO S Scheduler Swift Adapter 1 -API (new) Gateway 2 Rack 1 Server 1 M/R Swift Adapter 3 -Data (new) 4 Rack 2 Server 2 M/R Swift Adapter -Data (new) OSD OSD OSD OSD OSD OSD Gateway 1. Scheduler ask service where a particular block locates (control path) If the data already existed in one of the gateway, - API returns the location If the data not existed, - API returns a random gateway 2. Scheduler allocates a task on the server that is near to the data 3. Task access data from nearby (data path) 4. -Data relays the request to the backend if necessary (data path) Change to the existing is required! The rest remains unchanged.

33 Rack 1 Server 1 Rack 2 Server 2 Servic e Scheduler New Adapter 1 Master (NEW!) Gateway 2 M/R New Adapter 3 (modified) 4 Gateway M/R New Adapter (modified) 1. Scheduler ask service where a particular block locates (control path) If the data already existed in one of the gateway, - API returns the location If the data not existed, - API returns a random gateway 2. Scheduler allocates a task on the server that is near to the data 3. Task access data from nearby (data path) 4. -Data relays the request to the backend if necessary (data path) Ceph* RADO S OSD OSD OSD OSD OSD OSD

34 Rack 1 Server 1 Rack 2 Server 2 Servic e Scheduler Existing HDFS Adapter 1 HDFS Name Node Gateway 2 M/R Existing HDFS Adapter 3 HDFS Data Node (modified) 4 Gateway M/R Existing HDFS Adapter HDFS Data Node (modified) 1. Scheduler ask service where a particular block locates (control path) If the data already existed in one of the gateway, - API returns the location If the data not existed, - API returns a random gateway 2. Scheduler allocates a task on the server that is near to the data 3. Task access data from nearby (data path) 4. -Data relays the request to the backend if necessary (data path) Ceph* RADO S OSD OSD OSD OSD OSD OSD

35 Existing Existing implementation works in a different way and can t meet the requirement Client Client Load Balancer Servic e Share nothing Share nothing Ceph* RADO S OSD OSD OSD OSD OSD OSD

36 Servic e Ceph* RADO S Scheduler Adapter 1 - Proxy (NEW!) Gateway 2 Server 1 M/R Adapter (modified) OSD(SS D) Rack OSD(SS D) Rack 2 Server 2 M/R Adapter (modified) OSD OSD OSD OSD OSD OSD Gateway OSD(SS D) OSD(SS D) Cach e Tier Base Tier Sample Write(reducer) multiple uploads? Scheduler asks - Proxy for the location of a big data file -Proxy returns the closest instance Reducer writes the output to would split the object into multiple RADOS objects Ceph* CT would automatically handle the consistency here.

37 Testing Environment Host OS: CentOS7 Guest OS: CentOS7 Hadoop Nodes Cluster Baremetal OpenStack using KVM qemu-kvm v1.5 OpenStack using Docker Docker v1.6 37